1
$\begingroup$

I presently have 2 algorithms that have a numerical output. Using a threshold of 0.9, I get the classification output. Let's say they are:

  1. P (high precision, low recall)
  2. R (high recall, low precision)

Individually, they have poor F-1 scores. Is the naive way of creating a classifier C as:

C(*) = x.P(*) + (1-x).R(*)

And optimizing for x and threshold a good approach to improve the F-1 score? Or is there some alternate approach I must try. Note: I can't vary functions P() and R(). Their outputs are provided as a black-box function.

$\endgroup$
0
$\begingroup$

In general this would mean that P predicts only a small number of instances higher than 0.9 whereas R predicts most instances higher than 0.9. Therefore a weighted average of the two scores will fall somewhere in the middle, likely resulting in a moderate precision and moderate recall.

This can give significantly better results but only if the two classifiers are complementary, i.e. they predict instances in a way different enough from each other. Otherwise it's equivalent to tuning the threshold on a single classifier.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.